About Samuel

I lead applied AI work at ISx4, building production-oriented ML, LLM, agentic, and computer vision systems. My work sits between research depth and engineering delivery: turning ambiguous AI opportunities into reliable systems with evaluation, observability, security, and workflow fit built in from the start.

A few representative systems where the signal is in the delivery shape: translating ambiguous AI opportunities into systems teams can evaluate, operate, and improve. Some are private, internal, or client-facing, so I focus on the engineering pattern rather than repository links.

• 🏟 RunAI for GAA
  Applied AI for sports and organisational workflows, from product discovery through deployment-aware ML delivery.
  workflow modelling · applied ML · product discovery · operational fit

• 🤖 ISx4 ASQ -- Internal AI Assistant
  Enterprise assistant for ISx4 with governance, knowledge access, evaluation, and client-readiness built into the workflow.
  LLM systems · enterprise knowledge access · governance · evaluation

• 📡 AOP -- Agent Observability Platform
  Platform work for observing, evaluating, and improving AI agent behaviour in production-like environments.
  tracing · failure analysis · feedback loops · agent operations

• ✈ Customer Handling AI for Aviation Client
  Enterprise AI for aviation customer-handling workflows where safety awareness, reliability, and operational constraints matter.
  workflow automation · customer intelligence · reliability · operational constraints

Together, these projects reflect the work I enjoy most: moving from unclear AI opportunity to reliable system, with evaluation, monitoring, and workflow fit built in early.

How I tend to create value: by moving between research, systems thinking, product judgement, and production engineering without treating them as separate worlds.

• 🧠 Shape the AI opportunity
  Turn ambiguous ideas into scoped systems, delivery plans, evaluation criteria, and practical technical direction.
  discovery · architecture · technical leadership

• 🤖 Build LLM and agentic systems
  Design assistants, retrieval workflows, tool-using agents, governance paths, and feedback loops that teams can trust.
  RAG · agents · evaluation · governance

• 👁 Translate ML and computer vision research
  Bring statistical ML, deep learning, gaze/intention inference, and visual modelling into usable decision-support systems.
  computer vision · representation learning · human signals

• 🛠 Ship production-ready AI
  Build APIs, data pipelines, containers, monitoring, observability, and deployment paths that can survive real workflows.
  MLOps · observability · cloud · reliability

A few professional landmarks behind the systems work. CV available on request.

My research background sits behind the engineering: computer vision, dyadic interaction, gaze and visual cue modelling, human intention inference, and reliable ML. I use that depth to build AI systems that can handle noisy data, deployment constraints, and evaluation pressure.

I am also a Visiting Fellow at Queen's University Belfast, where my work connects machine learning, human-robot collaboration, and reliable visual inference.

Dyadic interaction, HRI, and intention inference

• Deep learning of dyadic interaction visual cues for human-robot collaboration in assembly tasks
  PhD thesis, Queen's University Belfast, 2024. Dyadic interaction, visual cues, gaze estimation, task recognition, action recognition, and intention-aware human-robot collaboration.

• QUB-PHEO: A Visual-Based Dyadic Multi-View Dataset for Intention Inference in Collaborative Assembly
  IEEE Access, 2024. Multi-view dyadic interaction dataset for intention inference in collaborative assembly.
  Dataset · arXiv

• Hand-Eye-Object Tracking for Human Intention Inference
  IFAC-PapersOnLine, 2022. Intention inference from hand movement, eye fixation, and object interaction cues.

• Dyadic Human-Robot Interaction: Emerging Technologies, Challenges, and Opportunities
  Book chapter, 2025. A broader treatment of dyadic HRI, emerging technologies, and open challenges.

• Establishing Baselines for Dyadic Visual Motion Prediction Using the QUB-PHEO Dataset
  IFAC-PapersOnLine, 2025. Reproducible baselines for motion prediction on QUB-PHEO.

Applied computer vision and reliable ML

• SLYKLatent: A Learning Framework for Gaze Estimation Using Deep Facial Feature Learning
  IEEE Transactions on Human-Machine Systems, 2025. Gaze estimation and facial feature representation learning.
  arXiv

• AlzhiNet: Traversing from 2D-CNN to 3D-CNN, Towards Early Detection and Diagnosis of Alzheimer's Disease
  Interdisciplinary Sciences: Computational Life Sciences, 2026. Hybrid 2D/3D CNN representations for Alzheimer's disease diagnosis.
  arXiv

• ISC-Perception: A Hybrid Computer Vision Dataset for Object Detection in Novel Steel Assembly
  arXiv, 2025. Hybrid synthetic and real-world perception dataset for intermeshed steel connection assembly.

• A Proposed Strategy for Automating Intermeshed Steel Connection Assembly using Robotics
  ISARC, 2025. Robotic assembly strategy for intermeshed steel connections.

• Application of Deep Learning to Autonomous Robotic Car
  International Journal of Computer Applications, 2021. Computer vision for autonomous robotic-car perception.

Datasets, tools, and reproducible research artifacts

• QUB-Perception of Human Engagement in Assembly Operation Dataset
  Zenodo dataset release for PHEO/QUB-PHEO research.

• Preprocessing Repository of QUB-Perception of Human Engagement in Assembly Operations Dataset
  Reproducible preprocessing artifact supporting the QUB-PHEO dataset.

• QUBVidCalib: Video Calibration and Correction Toolbox
  Calibration and correction tooling for multi-view video workflows.

• aVerify: A Video Annotation Verification Tool
  Tooling for validating and checking video annotation quality.

• Ormedian-Utils: A Computer Vision Utilities Package
  Utility package for computer vision workflows.

Technical notes

• Camera Calibration Demystified: Part 2 - Applications and Lens Distortion
  Practical camera calibration notes for robotics, autonomous systems, and lens distortion.

• Understanding Principal Component Analysis (PCA): A Comprehensive Guide
  Mathematical and code-oriented guide to PCA, dimensionality reduction, and practical ML use cases.

For the full publication list, see my Google Scholar.

• Enterprise AI agents with evaluation, observability, and clear operating boundaries
• LLM application quality: testing, monitoring, retrieval, and feedback loops
• Computer vision in high-stakes settings
• Production ML architecture across data, model, service, and user workflows
• Data-centric AI practices that turn usage and feedback into better systems

Open to conversations around AI Engineering, Applied AI Leadership, ML Engineering, Principal Data Scientist, Principal AI Engineer, and Applied Research roles where research depth and production delivery both matter.